Self-taught engineer
I build tooling that makes autonomous AI agents verifiable, recoverable, and safe to run — a tamper-evident execution ledger, burn-after-reading secret handling, and a coding agent that keeps its receipts.
A causally-ordered, rewindable event-ledger for autonomous AI agents.
Every step an agent takes is recorded in a hash-chained, tamper-evident ledger anyone can independently verify — then rewound to any prior state. Zero-trust by design, and no blockchain anywhere in sight.
Burn-after-reading secrets for AI agents.
Keys live in mlock'd, non-swappable RAM, are served to agents over MCP, and are destroyed after N reads — or the instant your session dies. A hardened daemon (seccomp allowlist, no core dumps, a non-elidable memory wipe) shipped with a fully-documented, honest threat model.
A verifiable AI coding agent that keeps the receipts.
Works with any model and keeps a checkable record of every change it makes — the agent loop bolted to korg's verification guarantees. Published on PyPI.
The local execution and self-recovery substrate beneath korgex: runs code fast and repairs its own errors.
Turn any website into agent-callable MCP tools with one line of config. Live at wmcp.sh.
Click any repository and get a straight answer on whether to actually use it.
Detects an LLM contradicting its own stated probabilities and prices the slip as a guaranteed-loss bet.
A deterministic semantic execution kernel with a governance-first architecture.
The full set of public repositories on GitHub.
Ledgers, execution substrates, and verification layers that make autonomous agents auditable and recoverable rather than opaque.
Low-level, correctness-first systems work — locked memory, syscall sandboxing, deterministic state machines, hash-chained data structures.
Secret handling, threat modeling that states its own limits honestly, and tooling that shrinks the window plaintext is reachable.